Alibaba has made its open-source Mixture-of-Experts model Qwen3.8-2.4T-A95B
available on its supercomputing network. The model is deployed on core nodes of
a 10,000-card supercluster and is distributed as an image through Alibaba's AI
community, built on Zhongzhi's FlagOS unified software stack and adapted for
domestic heterogeneous accelerator cards to support rapid enterprise and
developer deployment of business AI services.